Material and Methods

Questionnaire design and dissemination

A questionnaire was developed to collect information on French pet cats regarding their personality (five personality traits model) and the frequency with which their owners observed them bringing home birds and small mammals (ranging from never to very often, defined as once a week or more). To control for potential confounding factors, additional information was gathered on other characteristics of the cats (sex, age, and breed) and their living conditions (type of dwelling, type of the area around the dwelling, amount of time spent outdoors). The questionnaire was hosted online on the Google Form platform.
Households in France with at least one pet cat were targeted through postings on social media. The survey was anonymous, and no personal information was collected from the respondents. In the introduction part of the survey, respondents indicated their consent to participate in the study. The study complied with the legal requirements in France: as no personal information was collected, ethics approval was not mandatory, as was confirmed by the Research Ethics Committee of Paris-Saclay (Polethis, report from January 4, 2021).
The questionnaire consisted of four sections (Appendix 1). The first section focused on the cat characteristics: sex (female, male, unknown), age (<1 years, 1-2 years, 2-10 years, over 10 years, unknown), breed (Bengal, Birman, British Shorthair, Chartreux, Maine Coon, Persian, Ragdoll, Savannah, Sphynx, Siamese, Turkish Angora, non-pedigree, European, other, unknown). Note that in France “European” means “non-pedigree.” We offered the two options, because some owners might have been more familiar with one word than the other. The second section focused on the living conditions of the cat: type of housing (apartment without balcony, apartment with balcony, subdivision house, individual house), type of environment (urban, suburban, or rural area), time spent by the cat outdoors daily (none, limited [less than 1 hour], moderate [1 to 5 hours], long [more than 5 hours], all the time [just comes back to eat]), daily time spent by the owner with the cat (none, limited [less than 1 hour], moderate [1 to 5 hours], long [more than 5 hours]).
The third section involved assessing the personality traits. Litchfield et al. (2017) determined that the personality profiles of cats are organized around five factors that represent traits related to neuroticism, extraversion, agreeableness, dominance, and impulsivity. Each factor can be evaluated using a list of adjectives that have varying correlations with the trait in question. In our questionnaire, to ensure short response times and thus high completion rates (e.g., Plowman et al., 2013), we selected 15 of the 52 adjectives used in the original study of Litchfield et al. (2017). For each of the five personality traits, we selected three adjectives based on two criteria. First, we selected adjectives with unequivocal translations in French to avoid ambiguity for the respondents. Second, we used the factor scores of Litchfield et al. (2017) for each adjective to select those with a high correlation with the relevant personality trait and a low correlation with the four others in order to facilitate the interpretation of the results. For example, the adjective affectionate was selected, because it is readily translatable in French and has a high correlation with the personality trait of agreeableness and a low correlation with the four other personality traits. In English, we chose the following 15 adjectives. For neuroticism we chose: shy, calm (negative loading), and fearful of other cats; for extraversion: smart, vigilant, and persevering; for agreeableness: affectionate, friendly to people, and solitary (negative loading); for dominance: bullying, dominant, and aggressive to other cats; and for impulsiveness: impulsive, predictable (negative loading), and distractible. Each of these 15 adjectives was presented to the owners who could choose between strongly disagree, disagree, neither agree nor disagree, agree, and strongly agree.
The final section of the questionnaire focused on the prey returned home as observed by the owners and included the reported frequency of return of birds and small mammals (daily, 1-6 times per week, 2-3 times per month, 1-3 times per trimester, 1-3 times per year, never). At the end of the survey, we added an optional open-ended question to give the owners the opportunity to add any comments that they wished to share about their cat and to communicate any insights that may have been overlooked in the survey (Harland et al., 2011).
The social network Facebookwas chosen as a channel to disseminate the questionnaire. This network has a large number of user groups dedicated to cats, which made it possible to conduct a large-scale study. The questionnaire was diffused in 23 French-speaking groups from February 9, 2021, to March 14, 2021, which allowed us to collect a total of 3,217 responses. The dataset was deposited in the Mendeley repository [ref number XX–not available yet].

Data treatment and statistical analyses

All analyses were conducted using R v.4.2 (http://www.R-project.org). For all statistical tests, the level of significance was set at 0.05.
Factor analysis of personality structure
To ensure that the personality traits of the cats were reliably described by the owners, we removed from the analyses the surveys indicating that the owners did not spend any time with their cat (n = 10) and those with comments that prevented their use to study the personality traits (e.g., respondent not the cat owner, recently adopted cat, cat with a major health issue; n = 21). Like all animal species, cats go through different stages of development, and in juveniles, personality and predatory behavior are not yet stable (Lowe et al., 2001). Following Litchfield et al. (2017), we conservatively excluded cats aged under 1 year (n = 614) from the dataset. Finally, the responses with missing data in the set of personality adjectives were also removed (n = 64). The final dataset included 2,508 responses.
We performed exploratory factor analysis on the personality variables (15 adjectives evaluated by the owners) to first determine the number of personality traits to be extracted and then estimate the values of each trait for each cat (Sofroniou et al., 1999). We initially ensured that the data were suitable for factor analysis using Bartlett’s test of sphericity, which was significant (P < 0.01), and the Kaiser-Meyer-Olkin (KMO) criterion, which had an overall value of 0.722 (depending on the authors, a value above 0.5 or 0.6 is considered to mean that the sampling is adequate). Both indicators thus show that the data were suitable for factor analysis (Kaiser, 1974; Sofroniou et al., 1999). To determine the number of factors – here, personality traits – to extract, we used the empirical Kaiser criterion (see Braeken et al., 2017 for details) and parallel analysis with principal component analysis (PCA) (comparing the eigenvalues obtained to those generated from a Monte-Carlo simulated matrix), which indicated that four factors should be retained. We therefore choose to retain four factors, and used maximum likelihood factor analysis to extract them from the 15 adjectives. We obliquely rotated these factors (i.e., correlations between them were allowed), because personality traits are frequently correlated, as shown by the existence of personality syndromes. As in previous studies (e.g., Weiss et al., 2015), to interpret the factors, we defined salient loadings as those equal to or greater than |.4|. The four factors that we obtained correspond approximately a posteriori to Litchfield et al.’s (2017) extraversion (MR1), dominance (MR2), neuroticism (MR3), and agreeableness (MR4) (see Results section below).
Breed and personality
We tested the link between breed and personality traits by computing the Euclidian distance between personality profiles for each pair of cats to produce a resemblance matrix from which we conducted a nonparametric (permutational) analysis of variance (permanova; package vegan; Oksanen et al., 2007) using 999 permutations to test whether personality profiles differed according to breed. We then performed discriminant analysis using a non-parametric version of Pillai’s test to evaluate the significance of the eigenvalues (package ade4; Dray et al., 2007). To ensure to have enough statistical power, we removed five breeds with less than 60 individuals (Chartreux, n = 28; Savanah, n = 20; Sphinx, n = 57, Siamese, n = 49, Turkish Angora, n = 21), leading to a dataset of 2,162 responses. We regrouped the “Non-pedigree” and “European” cats under the “Non-pedigree” label, as both terms are used to describe the same type of cat in France (the European Short Hair breed exists but is extremely rare in France).
Factors influencing owner-reported frequency of prey brought home
To run the subsequent analyses regarding predation, additional responses were excluded from the previously described dataset: cats living in apartments with minimal outdoor access as well as cats living in houses but without daily outdoor access (n = 1217), owners of four or more cats (Cordonnier et al., 2022) who would supposedly have difficulty determining which cat brought home which prey (n = 188), incomplete responses (n = 2), and cats belonging to the Bengal breed (n = 36) because when the survey was posted on a Bengal cat Facebook group, several people suggested in the comments that participants give false answers to questions relating to predation. The final dataset included 719 responses. Since the response variables for reported predation frequencies were ordered (0: never, 1: 1 to 5 times a year, 2: 5 to 10 times a year, 3: 1 to 3 times a month; 4: once a week or more), two cumulative logit models (CLMs) were adjusted (McCullagh, 1980), with the reported frequencies of predation taken as response variables: CLM1: birds, CLM2: mammals. Each model incorporated 11 predictor variables: cat sex (0: male, 1: female), cat age (ordinal variable: 0: 1-2 years, 1: 2-10 years, 2: 10 years and older), cat breed (Birman, British Shorthair, Maine coon, Persian, Ragdoll, non-pedigree), type of housing (0: in a subdividion, 1: detached home), abundance of natural elements (tree, bushes, grass, etc.) around the place of residence (0: low, 1: moderate, 2: high), urban level (rural, suburban, urban), time spent daily outdoors (0: limited, 1: moderate, 2: long, 3: all the time), and the four quantitative personality traits (MR1 to MR4). To avoid potential multicollinearity issues, we ensured that all Variance Inflated Factors (VIFs) were under the threshold of 2 (O’Brien, 2007). The two models were fitted using the package ordinal (Christensen, 2015). For both models, we used a stepwise selection by sequential replacement to identify the subset of variables in the dataset resulting in the best performing model with the lowest prediction error (Venables et al., 2002; Hegyi et al., 2015). Wald tests were performed on the predictor variables. The quality of the model estimates was monitored using Pearson residuals (package sure; Greenwell et al., 2018). For the qualitative variables, post-hoc tests (including a Holm correction) were performed using a self-designed contrast matrix (package lsmeans; Lenth et al., 2016).